The revolution in protein structure prediction brought about by AlphaFold and AlphaFold 2 has opened new research possibilities in structural biology. Though the AlphaFold database (consisting of 214 million protein structures) is open source, its unstructured and voluminous nature presents a significant challenge for extracting meaningful knowledge. As a result, many approaches dedicated to compressing raw protein structure files (.mmcif or .pdb) into vectors have emerged. However, the quality of these embedding methods, i.e., how well they preserve the information found in raw data in the latent space of the embeddings, may vary dramatically. There is no standardized benchmark, evaluation pipeline, or comprehensive comparison of these methods. The goal of this thesis is to design and implement an evaluation framework for assessing the quality of available protein embedding methods for the whole (214M) AlphaFold Database. The student will: 1) Conduct extensive research into available protein embedding methods and related techniques to establish a foundational understanding of the current state of the field, 2) Select a representative subset of the embedding methods to be computed on the entire AlphaFold DB and design a computing pipeline on e-infra infrastructural resources, 3) Design and implement an evaluation framework capable of assessing the quality of embedding methods. In particular, the student will focus on analyzing the preservation of similarity relationships between protein structures in the raw form (typically used metric is the TM-Score) to the latent space of the vector embeddings (typically used metrics are Euclidean distance and Cosine distance). The student will explore the preservation of distance orderings using a suitable set of correlation and classification metrics. 4) Execute the evaluation framework using the e-infra computing infrastructure on the selected set of embedding methods for a representative set of protein structures, and 5) Perform a comparative analysis of the performance of the embedding methods. Following the principles of Open Science, the thesis' output will also be open-sourced artifacts, i.e., the computed embeddings of the selected methods in a suitable data repository, and the evaluation framework in a code repository.